services/nodemanager/tests/test_computenode_dispatch_slurm.py

   1 #!/usr/bin/env python
   2 # Copyright (C) The Arvados Authors. All rights reserved.
   3 #
   4 # SPDX-License-Identifier: AGPL-3.0
   5
   6 from __future__ import absolute_import, print_function
   7
   8 import subprocess
   9 import time
  10 import unittest
  11
  12 import mock
  13
  14 import arvnodeman.computenode.dispatch.slurm as slurm_dispatch
  15 from . import testutil
  16 from .test_computenode_dispatch import ComputeNodeShutdownActorMixin, ComputeNodeUpdateActorTestCase
  17
  18 @mock.patch('subprocess.check_output')
  19 class SLURMComputeNodeShutdownActorTestCase(ComputeNodeShutdownActorMixin,
  20                                             unittest.TestCase):
  21     ACTOR_CLASS = slurm_dispatch.ComputeNodeShutdownActor
  22
  23     def check_slurm_got_args(self, proc_mock, *args):
  24         self.assertTrue(proc_mock.called)
  25         slurm_cmd = proc_mock.call_args[0][0]
  26         for s in args:
  27             self.assertIn(s, slurm_cmd)
  28
  29     def check_success_after_reset(self, proc_mock, end_state='drain\n', timer=False):
  30         self.make_mocks(arvados_node=testutil.arvados_node_mock(63))
  31         if not timer:
  32             self.timer = testutil.MockTimer(False)
  33         self.make_actor()
  34         self.check_success_flag(None, 0)
  35         # At this point, 1st try should have happened.
  36
  37         self.timer.deliver()
  38         self.check_success_flag(None, 0)
  39         # At this point, 2nd try should have happened.
  40
  41         # Order is critical here: if the mock gets called when no return value
  42         # or side effect is set, we may invoke a real subprocess.
  43         proc_mock.return_value = end_state
  44         proc_mock.side_effect = None
  45
  46         # 3rd try
  47         self.timer.deliver()
  48
  49         self.check_success_flag(True, 3)
  50         self.check_slurm_got_args(proc_mock, 'NodeName=compute63')
  51
  52     def make_wait_state_test(start_state='drng\n', end_state='drain\n'):
  53         def test(self, proc_mock):
  54             proc_mock.return_value = start_state
  55             self.check_success_after_reset(proc_mock, end_state)
  56         return test
  57
  58     for wait_state in ['alloc\n', 'drng\n']:
  59         locals()['test_wait_while_' + wait_state.strip()
  60                  ] = make_wait_state_test(start_state=wait_state)
  61
  62     for end_state in ['idle*\n', 'down\n', 'down*\n', 'drain\n', 'fail\n']:
  63         locals()['test_wait_until_' + end_state.strip()
  64                  ] = make_wait_state_test(end_state=end_state)
  65
  66     def test_retry_failed_slurm_calls(self, proc_mock):
  67         proc_mock.side_effect = subprocess.CalledProcessError(1, ["mock"])
  68         self.check_success_after_reset(proc_mock)
  69
  70     def test_slurm_bypassed_when_no_arvados_node(self, proc_mock):
  71         # Test we correctly handle a node that failed to bootstrap.
  72         proc_mock.return_value = 'down\n'
  73         self.make_actor(start_time=0)
  74         self.check_success_flag(True)
  75         self.assertFalse(proc_mock.called)
  76
  77     def test_node_undrained_when_shutdown_cancelled(self, proc_mock):
  78         try:
  79             proc_mock.side_effect = iter(['', 'drng\n', 'drng\n', ''])
  80             self.make_mocks(arvados_node=testutil.arvados_node_mock(job_uuid=True))
  81             self.timer = testutil.MockTimer(False)
  82             self.make_actor()
  83             self.shutdown_actor.ping()
  84             self.busywait(lambda: proc_mock.call_args is not None)
  85             self.shutdown_actor.cancel_shutdown("test")
  86             self.check_success_flag(False, 2)
  87             self.assertEqual(proc_mock.call_args_list,
  88                              [mock.call(['scontrol', 'update', 'NodeName=compute99', 'State=DRAIN', 'Reason=Node Manager shutdown']),
  89                               mock.call(['sinfo', '--noheader', '-o', '%t', '-n', 'compute99']),
  90                               mock.call(['sinfo', '--noheader', '-o', '%t', '-n', 'compute99']),
  91                               mock.call(['scontrol', 'update', 'NodeName=compute99', 'State=RESUME'])])
  92         finally:
  93             self.shutdown_actor.actor_ref.stop()
  94
  95     def test_cancel_shutdown_retry(self, proc_mock):
  96         proc_mock.side_effect = iter([OSError, 'drain\n', OSError, 'idle\n', 'idle\n'])
  97         self.make_mocks(arvados_node=testutil.arvados_node_mock(job_uuid=True))
  98         self.make_actor()
  99         self.check_success_flag(False, 5)
 100
 101     def test_issue_slurm_drain_retry(self, proc_mock):
 102         proc_mock.side_effect = iter([OSError, OSError, 'drng\n', 'drain\n'])
 103         self.check_success_after_reset(proc_mock, timer=False)
 104
 105     def test_arvados_node_cleaned_after_shutdown(self, proc_mock):
 106         proc_mock.return_value = 'drain\n'
 107         super(SLURMComputeNodeShutdownActorTestCase,
 108               self).test_arvados_node_cleaned_after_shutdown()
 109
 110     def test_cancellable_shutdown(self, proc_mock):
 111         proc_mock.return_value = 'other\n'
 112         super(SLURMComputeNodeShutdownActorTestCase,
 113               self).test_cancellable_shutdown()
 114
 115     def test_uncancellable_shutdown(self, proc_mock):
 116         proc_mock.return_value = 'other\n'
 117         super(SLURMComputeNodeShutdownActorTestCase,
 118               self).test_uncancellable_shutdown()
 119
 120 @mock.patch('subprocess.check_output')
 121 class SLURMComputeNodeUpdateActorTestCase(ComputeNodeUpdateActorTestCase):
 122     ACTOR_CLASS = slurm_dispatch.ComputeNodeUpdateActor
 123
 124     def test_update_node_weight(self, check_output):
 125         self.make_actor()
 126         cloud_node = testutil.cloud_node_mock()
 127         arv_node = testutil.arvados_node_mock()
 128         self.updater.sync_node(cloud_node, arv_node).get(self.TIMEOUT)
 129         check_output.assert_called_with(['scontrol', 'update', 'NodeName=compute99', 'Weight=99000'])